FinSA Meeting Minutes with AI: Compliant in Minutes

Under the Swiss Financial Services Act (FinSA, in German FIDLEG) you must document your advice. Nobody says you have to type it up. AI can turn a meeting recording or your bullet points into a structured protocol in your fixed template: obtain consent, record or note, transcribe, summarise, review, approve, file. In my projects, the effort for a clean protocol drops by 90 to 95 percent. Twenty minutes of follow-up work become two minutes of review.

Everyone who advises clients knows the starting point. A client meeting takes an hour. The documentation afterwards costs another large block, usually in the evening, usually from memory, usually with the quiet doubt whether everything that might matter later is actually in there. Exactly this block is dispensable. The protocol itself is not.

What documentation does FinSA require?

Stated carefully, because the details depend on your activity and setup: the Financial Services Act (FIDLEG) and the associated Ordinance (FIDLEV) require, among other things, that financial service providers inform clients appropriately, check the suitability or appropriateness of recommendations and document their services (Art. 15 FinSA). For investment advice there is an additional duty to record the client's needs and the reasons for each recommendation. In practice this means: what you explained, recommended and agreed with the client must be demonstrable later.

What exactly that covers in your case, at what level of detail and for which clients, is a question for your compliance specialist, not for a blog article. What a blog article can answer: how the documentation comes into being without eating your evening. Because the obligation concerns the result, not the typing.

How does AI-based protocol creation work?

The workflow has six stations, and only one of them is new:

1. Consent. Before any recording runs, you obtain the client's consent, briefly, understandably, documented. Anyone who does not want a recording gets the classic variant: your bullet points work as a basis for the AI as well.
2. Recording or notes. The meeting is recorded, or you take bullet points as before. Nothing about the meeting itself changes.
3. Transcript. The recording becomes a verbatim transcript. It is working material, not the protocol.
4. Summary in a fixed structure. The AI fills your protocol template: the client's concern, options discussed, risks explained, agreements, next steps. Fixed structure means: every protocol in the firm looks the same and is complete.
5. Review and approval by the advisor. You read the draft against your memory, correct, add and approve. That takes minutes, because you review instead of formulating. Nothing leaves the system without this approval.
6. Filing. The approved protocol lands versioned in the client dossier; the transcript and the recording follow your deletion periods.

The decisive point sits in station five: responsibility for the content stays entirely with the advisor. The AI drafts, the human decides what is correct.

What does a good protocol structure look like?

The template that has proven itself in practice is short and always the same:

• Date, participants, occasion of the meeting
• The client's concern and situation in their own words
• Options discussed and recommendations made
• Risks explained and notices given
• Agreements and next steps with dates
• Reference to the consent for recording

One possible protocol structure: short, fixed, complete.

Because the structure is fixed, a later review immediately shows if something is missing. And because the AI pours every meeting into the same structure, you get the consistency that handwritten notes across five advisors never achieve.

What applies to data protection and retention?

Three rules suffice as a framework:

• Consent first. No recording without documented consent. The consent belongs in the protocol, not in an email thread.
• Define the storage location. Meeting recordings and transcripts contain client data and belong in a protected environment with a clear contract and a verified location, not in public chatbots. The revised Swiss Data Protection Act sets out the principles; the FDPIC publishes understandable factsheets (edoeb.admin.ch). What is allowed with client data and what is not is covered in the article on ChatGPT in a wealth manager's daily work.
• Define deletion periods. The approved protocol is retained. Whether and how long the raw recording and transcript are kept is something you define once and then follow. Many firms delete the raw data after the protocol is approved, because the protocol is the legally relevant document. That, too, is a question for your specialist.

How do the large institutions do it?

How advisor assistants prepare and follow up meetings at scale is something I see up close at additiv. An advisor assistant, launched at one of the world's largest insurers, goes further than the protocol: it suggests next best actions based on client history, prepares contracts and documents completely before the meeting and enriches the CRM via its voice function with soft client data, such as hobbies or pets, which makes segmentation and campaigns possible in the first place. The goal there is the same as here: more and better time with the client, because the paperwork disappears.

The SME version of this is deliberately smaller: no platform project, no CRM integration on day one. A recording or your bullet points, a fixed template, a protected storage location, your approval. That is enough to capture the 90 to 95 percent. Everything else can grow, but it does not have to.

From my practice: The most common objection does not come from compliance, but from experienced advisors: "I have written my own protocols for twenty years." My answer: exactly. Someone who has written protocols for twenty years can tell in two minutes whether a draft is correct. That ability to review is the real bottleneck, and it remains fully intact. What disappears is the typing that nobody ever learned because it interested them.

A separate legal framework applies to insurance brokers

Same logic, different legal basis: for insurance intermediaries it is the revised Insurance Supervision Act (VAG) and the Supervision Ordinance that have tightened information and documentation duties.

Note: We are not a law firm. All statements on legal and regulatory topics, in particular on FinSA/FIDLEG, FIDLEV and the revised Data Protection Act, are subject to change and do not replace legal advice. For specific questions, please consult a lawyer or your compliance specialist.

Conclusion: your next step

Take the next client meeting and afterwards drop your bullet points into the AI template instead of writing out the protocol. Review the draft and time yourself. If the result is right, and it will be once your template is clean, you have understood the process and can roll it out across the firm. Where protocols sit in the overall order of use cases is shown in the pillar article on AI for independent wealth managers. And what an environment looks like in which consent, storage location and approval are built in from the start: AI for wealth managers.

FAQ

Is recording a client meeting allowed under data protection law?

With documented client consent and a clarified storage location it is generally possible. The details, such as how long the raw recording is kept, should be clarified with your data protection or compliance specialist. The alternative without recording remains: your bullet points are a sufficient basis for the AI.

Does an AI-generated protocol meet FinSA requirements?

FinSA requires documentation, not a particular method of creating it. What matters is that the content is correct and demonstrable: what you informed, recommended and agreed. That is why human review and approval are a fixed part of the workflow. Interpretation in the individual case belongs with your compliance specialist.

What happens if the AI misunderstands something?

That is what station five is for: you review every draft before it enters the dossier. Experienced advisors spot errors in minutes, because they know what was said in the meeting. What the AI does not know, it does not invent: without content in the recording, the field stays empty, and you see exactly that during review.

Do we need a recording, or are notes enough?

Both work. A recording delivers the more complete result because nothing is lost. Bullet points are enough for a solid summary if they contain the core elements: concern, options, risks, agreements. Many firms start with notes and switch to recordings later.

What does getting started cost?

Expect a setup in the five-figure range and modest monthly running costs; the exact figures depend on your setup. Details are in the article What does AI cost for financial service providers?. Protocols are among the use cases with the fastest effect, because they occur daily.